An unconstrained multiphase thresholding approach for image segmentation
|
|
- Eustacia O’Neal’
- 5 years ago
- Views:
Transcription
1 An unconstrained multiphase thresholding approach for image segmentation Benjamin Berkels Institut für Numerische Simulation, Rheinische Friedrich-Wilhelms-Universität Bonn, Nussallee 15, Bonn, Germany WWW home page: Abstract. In this paper we provide a method to find global minimizers of certain non-convex 2-phase image segmentation problems. This is achieved by formulating a convex minimization problem whose minimizers are also minimizers of the initial non-convex segmentation problem, similar to the approach proposed by Nikolova, Esedoḡlu and Chan. The key difference to the latter model is that the new model does not involve any constraint in the convex formulation that needs to be respected when minimizing the convex functional, neither explicitly nor by an artificial penalty term. This approach is related to recent results by Chambolle. Eliminating the constraint considerably simplifies the computational difficulties, and even a straightforward gradient descent scheme leads to a reliable computation of the global minimizer. Furthermore, the model is extended to multiphase segmentation along the lines of Vese and Chan. Numerical results of the model applied to the classical piecewise constant Mumford-Shah functional for two, four and eight phase segmentation are shown. 1 Introduction Image segmentation is one of the fundamental research topics in the field of image processing. In particular, the Mumford-Shah model [1] is widely used in this context. One of the difficulties of this and many other variational image processing models is that the underlying energy functional has local, non-global minima. This is not only a theoretical problem, since the commonly used numerical minimization techniques often get stuck in local minima that differ considerably from a global minimum, hence possibly producing useless results. The goal of this paper is to introduce a method to obtain a global minimizer of the Mumford-Shah functional for 2-phase segmentation that only involves solving an unconstrained convex minimization problem. This method can be extended to multiphase segmentation by the ideas of Vese and Chan [2] in a canonical way. 1.1 Related work The problem of minimizing the Mumford-Shah segmentation functional has been extensively studied in the last decade leading to a wide range of existing methods,
2 each with its own shortcomings. One of the first numerical feasible methods to obtain (local) minimizers of the functional was proposed by Chan and Vese [3]. They build on the levelset methods of Osher and Sethian [4] and parameterize the unknown set by a levelset function. Shen [5] developed a Γ -convergence formulation along with a simple implementation by the iterated integration of a linear Poisson equation. The unknown set is represented in a diffuse way by a phase field. In [6], Esedoḡlu and Tsai tackle the minimization problem based on the threshold dynamics of Merriman, Bence and Osher [7] for evolving an interface by its mean curvature. Here the minimization is achieved by alternating the solution of a linear parabolic partial differential equation and simple thresholding. Alvino and Yezzi [8] approximate Mumford-Shah segmentation using reduced image bases. According to them, the majority of the robustness of Mumford-Shah segmentation can be obtained without allowing each pixel to vary independently. Their approximative model has comparable performance to Mumford-Shah segmentations where each pixel is allowed to vary freely. A way to obtain global minimizers was introduced by Nikolova, Esedoḡlu and Chan [9]. Here, a convex constrained minimization problem has to be solved followed by a simple thresholding of the latter minimizer. This method is closely related to the method we propose in this paper, the key difference is that [9] requires a constraint in the convex minimization while the model proposed in this paper does not involve any constraint in the convex formulation. On the other hand there are methods to solve a certain class of minimal surface problems by unconstrained convex optimization, cf. the work of Chambolle and Darbon [10, 11]. The 2-phase Mumford-Shah functional belongs to this class, yet due to the best of our knowledge nobody seems to have tapped the potential offered by these general insights for Mumford-Shah based image segmentation so far. 2 Constrained global 2-phase minimization First let us describe the general framework and revise the work of Nikolova et al. [9], the starting point for our model. In the following, denotes our computational domain, an arbitrary but fixed subset of R n. For given indicator functions f 1, f 2 L 1 () such that f 1, f 2 0 a.e. we consider the prototype Mumford-Shah energy E MS [] := f 1 dx + f 2 dx + ν Per(), (1) where Per() denotes the perimeter of the set in. If u 0 is an image, c 1, c 2 R are two grey values and f i (x) := (u 0 (x) c i ) 2, this is the well known piecewise constant Mumford-Shah functional for 2-phase segmentation, i.e. E[, c 1, c 2 ] = (u 0 c 1 ) 2 dx + (u 0 c 2 ) 2 dx + ν Per(). (2) \ \
3 Remark 1. Because of E MS [] = (f 1 f 2 )dx + ν Per() + f 2 dx, }{{} =:ÊMS[] E MS and ÊMS share the same minimizers. Remark 2. For h(x) := e x 2, we have E MS [] = (f 1 + h) dx + \ (f 2 + h) dx + ν Per() hdx, }{{} =C< i.e. replacing f 1 and f 2 by f 1 + h and f 2 + h does not affect the minimizers of E MS. This, combined with f 1, f 2 0 a.e., means that we can assume f 1, f 2 > 0 a.e. in without loss of generality. To obtain (local) minimizers of the functional above, Chan and Vese [3] proposed to parametrize the unknown set by a levelset function φ and get the energy E CV [φ] := H(φ)f 1 + (1 H(φ))f 2 + ν (H(φ)) dx. Here, H( ) denotes the Heaviside function, i.e. H(s) = 1 for s > 0 and H(s) = 0 else. A gradient descent will be used for minimization, therefore H is replaced by a smeared out Heaviside function, e.g. H δ (x) := π arctan ( ) x δ, where δ > 0. While the specific choice is not important, it is important to use a function whose derivative does not have compact support (cf. [3]). This gives the regularized energy E CV,δ [φ] := H δ (φ)f 1 + (1 H δ (φ)) f 2 + ν (H δ (φ)) dx (3) and yields the gradient descent t φ = H δ(φ) [ ( )] φ (f 2 f 1 ) + ν div. (4) φ One of the major drawbacks of the energy (3) is its non-convexity in φ. In [9], Nikolova et al. noted that the gradient descent (4) and [ ( )] φ t φ = (f 2 f 1 ) + ν div φ have the same stationary points, because H δ (φ) > 0. Obviously the latter is the gradient descent of the energy E CE [φ] := (f 1 f 2 )φ + ν φ dx.
4 In general, f 1 f 2 takes positive and negative values, therefore the energy is not bounded (neither from below nor from above). In other words, it does not necessarily have a minimizer. However, this is easily fixed by restricting the minimization to 0 φ(x) 1 for all x. Based on this, the following theorem holds: Theorem 1. For given indicator functions f 1, f 2 L 1 () such that f 1, f 2 0 a.e., let u := argmin 0 ũ 1 (f 1 f 2 )ũ + ν ũ dx = argmin E CE [ũ] 0 ũ 1 and c := {x u(x) > c}. Then c is a minimizer of the Mumford-Shah energy (1) for all c [0, 1). Proof. Nikolova et al. proved this theorem in [9] for a.e. c [0, 1], we extend it here to hold not only for almost every, but for every c [0, 1). First, we briefly sketch the prove given by Nikolova et al. for a.e. c [0, 1]. Using 0 u 1 and the coarea formula, one can show E CE [u] = 1 0 E MS [ c ]dc C, where C is a constant independent of u. Let be a minimizer of E MS (the existence of such minimizers using convergence in measure follows from standard arguments) and let M := {c [0, 1] E MS [ c ] > E MS [ ]}. Assuming µ(m) > 0 leads to the contradiction E CE [χ ] < E CE [u] therefore µ(m) = 0 holds and the statement is proven for a.e. c [0, 1]. Here, χ A denotes the characteristic function of the set A. Now we extend the statement to all c [0, 1), inspired by the proof of Lemma 4 (iii) in [12]: Again let u be a minimizer of E CE under the constraint 0 u 1 and denote its superlevelsets by c. Choose an arbitrary but fixed ĉ [0, 1). The statement holds for a.e. c [0, 1], so by Remark 1, there exists a sequence (c n ) [0, 1] N with c n ĉ such that cn argmin Ê MS []. Since the superlevelsets of a function are contained in each other, we have χ cn = χ n χ k=1 c pointwise a.e., where := k n=1 c n. Setting g := f 1 f 2 and using Lebesgue s dominated convergence theorem, we obtain gdx = gχ dx = lim gχ cn dx = lim gdx. n n cn Here we used gχcn g f1 + f 2 to provide the integrable upper bound. For each n and, we have gdx + ν Per( cn ) gdx + ν Per(). cn
5 Using the continuity argument from above and the lower semicontinuity of the perimiter (cf. [13]), we get gdx + ν Per( ) gdx + ν Per(), i.e. is a minimizer of E MS [, c n ]. Combining this with c = {x u(x) > c} = {x u(x) > c n } = n=1 n=1 concludes the proof. cn Knowing that Theorem 1 holds true for all c [0, 1) also remedies the last bit of uncertainty left in [9]. Remark 3. For any function u that fulfills the constraint, obviously {u > 1} =. Therefore we cannot expect Theorem 1 to hold for c = 1. To solve the constrained optimization problem, Nikolova et al. show that the constrained problem has the same minimizers as the unconstrained problem if a penalty term of the form α p(u(x)) is added with a sufficiently large coefficient α (cf. [9], Claim 1). Here p denotes p(s) = max{0, 2 s 1 2 1}. While this result already gives a method to find global minimizers of E MS by solving a convex, unconstrained minimization problem, its practical relevance is limited. Most numerical minimizations methods rely on the gradient of the functional, but the proposed penalty term is not differentiable, making a regularization necessary. But any smooth regularization of the penalty term will stop the minimizers of the convex, constrained functional to coincide with those of the convex functional with penalty term. The stronger the regularization, the more the minimizers deviate. Furthermore, the regularization imposes numerical difficulties. If an explicit gradient descent is used for the minimization (as proposed in [9]), a suitable timestep size control is needed to ensure convergence. The step sizes allowed by such methods, e.g. the Armijo rule [14], typically correspond to the size of the region in which the linearization of the functional properly approximates the functional. Due to the nature of the penalty term p, the linearization at 0 and 1 of a regularized version of it only approximates the regularization properly in a region that is of the size of the regularization parameter. So, as soon as the current iterate of the gradient descent takes values near 0 or 1, the timestep control only allows timestep sizes of the order of the regularization parameter, which, as mentioned above, cannot be chosen too big. Instead of using a penalty term one could of course also approach the constrained convex optimization problem directly. This is done for example by Bresson et al. [15]. Their approach does not need a penalty term and gives an efficient algorithm to minimize E CE, but has to introduce an additional unknown v and a regularization parameter θ and needs to minimize for u and v alternatingly. Furthermore, the key idea to apply Chambolle s TV minimization algorithm [16] can
6 also be directly applied to our model to obtain a simpler and faster minimization algorithm: There is no need to introduce v, θ and the alternating minimization. Therefore it is worth to investigate whether it is possible to simplify the problem by getting rid of the constraint altogether. 3 Unconstrained global 2-phase minimization Another alternative to Chan Vese is a phase field approach [6, 5] with a typical double well term: ( ) 1 E PH,ɛ [u] := u 2 f 1 + (1 u) 2 f 2 + ν ɛ u2 (1 u) 2 + ɛ u 2 dx. A minimizer u ɛ of this energy is a diffuse representation of the segmentation, i.e. {u ɛ = 0} and {u ɛ = 1} represent the two segments respectively with a smooth transition in between. E PH,ɛ [u] is known to Γ -converge to E MS [5], but unfortunately not convex and does not permit jumps in u for ɛ > 0. Knowing both E CE and E PH,ɛ, the question arises whether it is possible to combine the advantages of both models while eliminating some of the disadvantages. Heuristically looking at both energies served as motivation to investigate the following energy: E[u] := u 2 f 1 + (1 u) 2 f 2 + ν u dx. (5) This energy is convex because it does not involve the non-convex double well term of E PH,ɛ, and can be minimized without imposing constraints because it does not have the indicator term from E CE that is not bounded from below. Furthermore, it permits jumps in u. Remark 4. Given a function u, obviously we have E[min{max{0, u}, 1}] E[u]. Therefore, a minimizer u min fulfills 0 u min 1. While the proposed functional has some nice obvious properties, it is far from obvious whether there is a relation between its minimizer and minimizers of E MS. Before we tackle this question, let us remark a link between E CE and E: Remark 5. There is a direct relationship between E CE and E: A straightforward calculation shows u 2 f 1 + (1 u) 2 f 2 = (f 1 f 2 ) u + (u 1 2 )2 (f 1 + f 2 ) 1 4 (f 1 + f 2 ) + f 2. Therefore E[u] = (f 1 f 2 ) u + (u 1 2 )2 (f 1 + f 2 ) 1 4 (f 1 + f 2 ) + f 2 + ν u dx =E CE [u] + (f 1 + f 2 )(u 1 2 )2 dx + C.
7 In other words, E essentially equals E CE plus an additional quadratic penalty energy. The constant C is clearly irrelevant for the minimizers. To investigate the relation between the minimizers of E and minimizers of E MS we can make use of the theory derived in the context of the connection between minimal surface problems and total variation minimization. The following general statement has been made by Chambolle [17], Chambolle and Darbon [11], in the continuous setting, its discrete counterpart is well known: Theorem 2. Let Ψ : R R, (x, s) Ψ(x, s) such that Ψ(x, ) is C 1 and uniformly convex for all x and u := argmin Ψ(x, ũ(x)) + ν ũ dx. ũ Then c := {x u(x) > c} for all c R is a minimizer of s Ψ(x, c)dx + ν Per(). Note that this general statement cannot be directly applied to the model of Nikolova et al. discussed in Section 2 because the integrand is neither uniformly (not even strictly) convex nor does the general statement incorporate the constraint. As remarked in [11], the proof for a more specific statement given in [10] still applies to Theorem 2. Theorem 3. If u is a minimizer of (5), then {u > 1 2 } minimizes E MS [] = f 1 dx + f 2 dx + ν Per(). \ Proof. Let Ψ(x, s) := s 2 f 1 (x)+(1 s) 2 f 2 (x). Obviously Ψ(x, ) is C 2 for all x and we have s Ψ(x, s) = 2sf 1 (x)+2(s 1)f 2 (x) and s 2 Ψ(x, s) = 2(f 1 (x)+f 2 (x)). From Remark 2, we know that f 1, f 2 > 0 a.e., therefore Ψ(x, ) is uniformly convex for a.e. x. Now just apply Theorem 2, noting s Ψ(x, 1 2 ) = f 1(x) f 2 (x) and Remark 1. In this sense, our theorem is a corollary of Theorem 2. The preceding theorem finally tells us how to find a global minimizer of E MS [ ] given in (1): Minimize the convex energy (5) and threshold the minimizer to 1 2. In case of the piecewise constant Mumford-Shah functional for 2-phase segmentation, we obtain a global minimizer of the Mumford-Shah energy (2) with respect to for fixed gray values c 1, c 2. We do not necessarily find a global minimizer with respect to, c 1 and c 2. Another link between E CE and E PH,ɛ is the so-called piecewise constant levelset method [18] for 2-phase segmentation that constrains the levelset function
8 to be piecewise constant. If this constraint is approximated with a penalty energy, the method equals the phase field approach. If the constraint is relaxed to a certain boundedness constraint, the method equals [9]. In both cases the fidelity term has to be altered accordingly, making use of the fact that this term is the same in E CE and E PH,ɛ if u only takes the values 0 and 1. Since (5) is similar to the Rudin-Osher-Fatemi energy [19], there is a wide variety of established minimization schemes to choose from, ranging from a straightforward gradient descent scheme with a differentiable approximation of the BV term over primal thresholding methods [20] to sophisticated methods based on the dual formulation of the BV norm, e.g. [16, 11]. With Ψ(x, s) = 1 2 (s (f 2(x) f 1 (x))) 2, another immediate consequence of Theorem 2 is that the zero superlevelset of a minimizer of the ROF energy E ROF [u] := 1 2 (u (f 2 f 1 )) 2 + ν u dx (6) is a global minimizer of ÊMS and therefore of E MS. This is another way to obtain a global minimizer of E MS by unconstrained convex optimization, but compared to (5) this method has a few shortcomings, cf. Sections 4 and 5. Furthermore, the boundedness mentioned in Remark 4 does not hold for minimizers of the ROF energy. Perhaps this is one of the reasons why nobody seems to have used the classical ROF function for Mumford-Shah based image segmentation so far. 4 Multiphase segmentation Our functional can be extended to multiphase segmentation by the using the idea of Vese and Chan [2] in a straightforward manner. To keep notation at bay, we restrict the discussion to segmentation in 4 phases. The segmentation in 2 n phases works analogously. Let f 1, f 2, f 3, f 4 L 1 () such that f i 0 a.e., then the multiphase functional is given by E[u 1, u 2 ] := u 2 1u 2 2f 1 + (1 u 1 ) 2 u 2 2f 2 + u 2 1(1 u 2 ) 2 f 3 + (1 u 1 ) 2 (1 u 2 ) 2 f 4 (7) + ν ( u 1 + u 2 ) dx. If we fix u 2, the reduced functional E[, u 2 ] is the same as the 2-phase functional (5) with the indicator functions f 1 = u 2 2f 1 + (1 u 2 ) 2 f 3 and f 2 = u 2 2f 2 + (1 u 2 ) 2 f 4. As in the 2-phase case, we can assume f i > 0 a.e. without loss of generality and because either u 2 2 > 0 or (1 u 2 ) 2 holds, we have f 1, f 2 > 0. Therefore, all statements proven for the 2-phase functional can be applied to E[, u 2 ], i.e. we can compute the global minimum (for fixed u 2 ). The same applies for fixed u 1, so as an optimization strategy, we propose to minimize with respect to u 1 and u 2 alternatingly. Even though it is easy to extend (5) to multiphase segmentation, the same does not apply to the ROF energy (6). There is no apparent extension in the sense of [2] to formulate the multiphase segmentation in a single functional.
9 5 Indicator parameters In typical segmentation tasks, the indicator functions depend on unknown parameters, e.g. the grey values for each segment in case of the piecewise constant Mumford-Shah model. For the sake of simplicity, we discuss the latter model in its 2-phase formulation here, i.e. f i (x) := (u 0 (x) c i ) 2, i = 1, 2, but this discussion applies to other indicator functions and multiphase segmentation as well. During the minimization of (5) we have to minimize for c 1 and c 2 as well. This is typically done in an alternating fashion, but there are two apparent possibilities to update the grey values: Minimize (5) with respect to c 1 and c 2 or do so for the energy in the set formulation (1). The two possible updating formulae for c 1 two arising are / / c 1 = u 2 u 0 dx u 2 dx or c 1 = u 0 dx dx. {u> 1 2 } {u> 1 2 } The two possibilities only coincide if u is binary. The first formula not only averages u 0 in {u > 1 2 }, instead it takes into account the values of u 0 everywhere, but weights the values according to u 2. To a certain degree this is similar to the effect of the regularization of the Heaviside function in the model of Chan and Vese. From our experiments, this reduces the chance of getting stuck in local minima that can still occur when minimizing over u and the indicator parameters. Particulary in the case of multiphase segmentation it turned out to be beneficial. Due to the different way f 1 and f 2 are used in the ROF energy (6), it is not quadratic in c 1 and c 2. So this functional does not give a natural formula to update the grey values. 6 Numerical examples To conclude, we show the practical usability of the proposed model by applying it to the classical piecewise constant Mumford-Shah functional, see equation (2). As minimization method we use an explicit gradient descent scheme with the Armijo rule [14] as timestep size control. The absolute value is regularized by z ɛ = z 2 + ɛ 2 (in all examples presented here, ɛ = 0.1 is used). For the spatial discretization, we use bilinear finite elements on a regular quadrilateral grid, i.e. each pixel of the input image u 0 corresponds to a node of the finite element mesh. The grey values c 1 and c 2 are initialized with 0 and 1 respectively and updated occasionally during the gradient descent. Figure 1 shows results of our method and of the one proposed by Nikolova et al. [9] on one artificial image and one digital photo. In both examples, the minimizer u from our model is far from being binary, but this is nothing to be expected from the theory presented in this paper. The 0.5-superlevelset gives an accurate segmentation that is not influenced by the presence of heavy noise (top row) and works on non-binary input images (bottom row). The minimizers u
10 Fig. 1. Segmentation of an artificial noisy structure (ν = , top row) and the well-known Matlab cameraman image (ν = , bottom row): Input image u 0 (left), segmentation function u and 0.5-superlevelset of u colored with the average grey values c 1, c 2 obtained by our model (middle) and by using E CE (right). The slight difference of the grey values is attributed to the employed update formula, cf. Section 5. of the Nikolova et al. model look very different, but the segmentation obtained from the 0.5-superlevelsets is almost identical. Upon closer inspection, the minimizer u of our model from the top row of Figure 1 looks very much like as obtained by minimizing the ROF energy with u 0 as input image. This is not surprising due to the following observation: If u 0 is binary, i.e. u 0 = χ A for a set A and c 1 = 0, c 2 = 1 we have f 1 = (χ A 0) 2 = χ A and f 2 = (χ A 1) 2 = χ \A and therefore E[u] = (u χ \A ) 2 + ν u dx, i.e. E equals the ROF energy in this special case. This is not the case if u 0 is non-binary which can be seen from the bottom row of Figure 1. Figure 2 shows 4-phase segmentation results. Those indicate the tendency of the segmentation functions to become binary for small values of ν. Finally, Figure 3 illustrates the behavior of the method for different numbers of segments and Figure 4 shows three timesteps of the 8-phase segmentation. References 1. Mumford, D., Shah, J.: Optimal approximation by piecewise smooth functions and associated variational problems. Communications on Pure Applied Mathematics 42 (1989) Vese, L.A., Chan, T.F.: A multiphase level set framework for image segmentation using the Mumford and Shah model. International Journal of Computer Vision 50(3) (2002) Chan, T.F., Vese, L.A.: Active contours without edges. IEEE Transactions on Image Processing 10(2) (2001)
11 Fig phase segmentation of an artificial noisy image (top row) and a MRI image (bottom row) (ν = ): Input image u 0 (left), segmentation functions u 1 and u 2 (middle), segmentation colored with the average grey values c 1,..., c 4 (right). 4. Osher, S.J., Sethian, J.A.: Fronts propagating with curvature dependent speed: Algorithms based on Hamilton Jacobi formulations. Journal of Computational Physics 79 (1988) Shen, J.: Γ -convergence approximation to piecewise constant Mumford-Shah segmentation. In: 7th International Conference on Advanced Concepts for Intelligent Vision Systems (ACIVS 05). Volume 3708 of LNCS. (2005) Esedoḡlu, S., Tsai, Y.H.R.: Threshold dynamics for the piecewise constant Mumford-Shah functional. Journal of Computational Physics 211(1) (2006) Merriman, B., Bence, J.K., Osher, S.J.: Diffusion generated motion by mean curvature. CAM Report 92-18, UCLA (1992) 8. Alvino, C.V., Yezzi, A.J.: Fast Mumford-Shah segmentation using image scale space bases. In: Society of Photo-Optical Instrumentation Engineers (SPIE) Conference Series. Volume (2007) 9. Nikolova, M., Esedoḡlu, S., Chan, T.F.: Algorithms for finding global minimizers of image segmentation and denoising models. SIAM Journal on Applied Mathematics 66(5) (2006) Chambolle, A.: An algorithm for mean curvature motion. Interfaces and free Boundaries 6 (2004) Chambolle, A., Darbon, J.: On total variation minimization and surface evolution using parametric maximum flows. CAM Report 08-19, UCLA (2008) 12. Alter, F., Caselles, V., Chambolle, A.: A characterization of convex calibrable sets in R N. Mathematische Annalen 332(2) (2005) Ambrosio, L., Fusco, N., Pallara, D.: Functions of bounded variation and free discontinuity problems. Oxford Mathematical Monographs. Oxford University Press, New York (2000)
12 Fig. 3. Segmentation of a digital photo (ν = ). Input image u 0 (left), segmentation in four (middle) and eight (right) segments colored with the average grey values of the segments. Original image c bigmama / PIXELIO. Fig. 4. Intermediate results of the segmentation in eight segments shown in Figure 3 after 50 (left), 250 (middle) and 700 (right) gradient descent steps. 14. Kosmol, P.: Methoden zur numerischen Behandlung nichtlinearer Gleichungen und Optimierungsaufgaben. 2. edn. Teubner, Stuttgart (1993) 15. Bresson, X., Esedoḡlu, S., Vandergheynst, P., Thiran, J., Osher, S.: Fast global minimization of the active contour/snake model. Journal of Mathematical Imaging and Vision 28(2) (2007) Chambolle, A.: An algorithm for total variation minimization and applications. Journal of Mathematical Imaging and Vision 20(1-2) (2004) Chambolle, A.: Total variation minimization and a class of binary MRF models. In: Energy Minimization Methods in Computer Vision and Pattern Recognition. Volume 3757 of LNCS. (2005) Lie, J., Lysaker, M., Tai, X.C.: A binary level set model and some applications to Mumford-Shah image segmentation. IEEE Transactions on Image Processing 15(5) (2006) Rudin, L., Osher, S., Fatemi, E.: Nonlinear total variation based noise removal algorithms. Physica D 60 (1992) Daubechies, I., Defrise, M., de Mol, C.: An iterative thresholding algorithm for linear inverse problems with a sparsity constraint. Communications on Pure and Applied Mathematics 57(11) (2004)
A posteriori error control for the binary Mumford Shah model
A posteriori error control for the binary Mumford Shah model Benjamin Berkels 1, Alexander Effland 2, Martin Rumpf 2 1 AICES Graduate School, RWTH Aachen University, Germany 2 Institute for Numerical Simulation,
More informationConvex Hodge Decomposition of Image Flows
Convex Hodge Decomposition of Image Flows Jing Yuan 1, Gabriele Steidl 2, Christoph Schnörr 1 1 Image and Pattern Analysis Group, Heidelberg Collaboratory for Image Processing, University of Heidelberg,
More informationConvex Hodge Decomposition and Regularization of Image Flows
Convex Hodge Decomposition and Regularization of Image Flows Jing Yuan, Christoph Schnörr, Gabriele Steidl April 14, 2008 Abstract The total variation (TV) measure is a key concept in the field of variational
More informationIMA Preprint Series # 2098
LEVEL SET BASED BIMODAL SEGMENTATION WITH STATIONARY GLOBAL MINIMUM By Suk-Ho Lee and Jin Keun Seo IMA Preprint Series # 9 ( February ) INSTITUTE FOR MATHEMATICS AND ITS APPLICATIONS UNIVERSITY OF MINNESOTA
More informationAnalysis of Numerical Methods for Level Set Based Image Segmentation
Analysis of Numerical Methods for Level Set Based Image Segmentation Björn Scheuermann and Bodo Rosenhahn Institut für Informationsverarbeitung Leibnitz Universität Hannover {scheuerm,rosenhahn}@tnt.uni-hannover.de
More informationENERGY METHODS IN IMAGE PROCESSING WITH EDGE ENHANCEMENT
ENERGY METHODS IN IMAGE PROCESSING WITH EDGE ENHANCEMENT PRASHANT ATHAVALE Abstract. Digital images are can be realized as L 2 (R 2 objects. Noise is introduced in a digital image due to various reasons.
More informationc 2006 Society for Industrial and Applied Mathematics
SIAM J. APPL. MATH. Vol. 66, No. 5, pp. 1632 1648 c 26 Society for Industrial and Applied Mathematics ALGORITHMS FOR FINING GLOBAL MINIMIZERS OF IMAGE SEGMENTATION AN ENOISING MOELS TONY F. CHAN, SELIM
More informationA Localized Linearized ROF Model for Surface Denoising
1 2 3 4 A Localized Linearized ROF Model for Surface Denoising Shingyu Leung August 7, 2008 5 Abstract 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 1 Introduction CT/MRI scan becomes a very
More informationA Variational Approach to Reconstructing Images Corrupted by Poisson Noise
J Math Imaging Vis c 27 Springer Science + Business Media, LLC. Manufactured in The Netherlands. DOI: 1.7/s1851-7-652-y A Variational Approach to Reconstructing Images Corrupted by Poisson Noise TRIET
More informationDual methods for the minimization of the total variation
1 / 30 Dual methods for the minimization of the total variation Rémy Abergel supervisor Lionel Moisan MAP5 - CNRS UMR 8145 Different Learning Seminar, LTCI Thursday 21st April 2016 2 / 30 Plan 1 Introduction
More informationA Pseudo-distance for Shape Priors in Level Set Segmentation
O. Faugeras, N. Paragios (Eds.), 2nd IEEE Workshop on Variational, Geometric and Level Set Methods in Computer Vision, Nice, 2003. A Pseudo-distance for Shape Priors in Level Set Segmentation Daniel Cremers
More informationA Level Set Based. Finite Element Algorithm. for Image Segmentation
A Level Set Based Finite Element Algorithm for Image Segmentation Michael Fried, IAM Universität Freiburg, Germany Image Segmentation Ω IR n rectangular domain, n = (1), 2, 3 u 0 : Ω [0, C max ] intensity
More informationA Convex Relaxation of the Ambrosio Tortorelli Elliptic Functionals for the Mumford Shah Functional
A Convex Relaxation of the Ambrosio Tortorelli Elliptic Functionals for the Mumford Shah Functional Youngwook Kee and Junmo Kim KAIST, South Korea Abstract In this paper, we revisit the phase-field approximation
More informationVariational Methods in Image Denoising
Variational Methods in Image Denoising Jamylle Carter Postdoctoral Fellow Mathematical Sciences Research Institute (MSRI) MSRI Workshop for Women in Mathematics: Introduction to Image Analysis 22 January
More informationNon Convex Minimization using Convex Relaxation Some Hints to Formulate Equivalent Convex Energies
Non Convex Minimization using Convex Relaxation Some Hints to Formulate Equivalent Convex Energies Mila Nikolova (CMLA, ENS Cachan, CNRS, France) SIAM Imaging Conference (IS14) Hong Kong Minitutorial:
More informationConvergence rates of convex variational regularization
INSTITUTE OF PHYSICS PUBLISHING Inverse Problems 20 (2004) 1411 1421 INVERSE PROBLEMS PII: S0266-5611(04)77358-X Convergence rates of convex variational regularization Martin Burger 1 and Stanley Osher
More informationAn Algorithm for Minimizing the Mumford-Shah Functional
An Algorithm for Minimizing the Mumford-Shah Functional Thomas Pock Graz University of Technology, Austria Horst Bischof Graz University of Technology, Austria Daniel Cremers University of Bonn, Germany
More informationNovel integro-differential equations in image processing and its applications
Novel integro-differential equations in image processing and its applications Prashant Athavale a and Eitan Tadmor b a Institute of Pure and Applied Mathematics, University of California, Los Angeles,
More informationAdaptive Primal Dual Optimization for Image Processing and Learning
Adaptive Primal Dual Optimization for Image Processing and Learning Tom Goldstein Rice University tag7@rice.edu Ernie Esser University of British Columbia eesser@eos.ubc.ca Richard Baraniuk Rice University
More informationA First Order Primal-Dual Algorithm for Nonconvex T V q Regularization
A First Order Primal-Dual Algorithm for Nonconvex T V q Regularization Thomas Möllenhoff, Evgeny Strekalovskiy, and Daniel Cremers TU Munich, Germany Abstract. We propose an efficient first order primal-dual
More informationarxiv: v1 [math.na] 3 Jan 2019
arxiv manuscript No. (will be inserted by the editor) A Finite Element Nonoverlapping Domain Decomposition Method with Lagrange Multipliers for the Dual Total Variation Minimizations Chang-Ock Lee Jongho
More informationA Dual Formulation of the TV-Stokes Algorithm for Image Denoising
A Dual Formulation of the TV-Stokes Algorithm for Image Denoising Christoffer A. Elo, Alexander Malyshev, and Talal Rahman Department of Mathematics, University of Bergen, Johannes Bruns gate 12, 5007
More informationAn image decomposition model using the total variation and the infinity Laplacian
An image decomposition model using the total variation and the inity Laplacian Christopher Elion a and Luminita A. Vese a a Department of Mathematics, University of California Los Angeles, 405 Hilgard
More informationImage Cartoon-Texture Decomposition and Feature Selection using the Total Variation Regularized L 1 Functional
Image Cartoon-Texture Decomposition and Feature Selection using the Total Variation Regularized L 1 Functional Wotao Yin 1, Donald Goldfarb 1, and Stanley Osher 2 1 Department of Industrial Engineering
More informationPDEs in Image Processing, Tutorials
PDEs in Image Processing, Tutorials Markus Grasmair Vienna, Winter Term 2010 2011 Direct Methods Let X be a topological space and R: X R {+ } some functional. following definitions: The mapping R is lower
More informationSTRUCTURAL OPTIMIZATION BY THE LEVEL SET METHOD
Shape optimization 1 G. Allaire STRUCTURAL OPTIMIZATION BY THE LEVEL SET METHOD G. ALLAIRE CMAP, Ecole Polytechnique with F. JOUVE and A.-M. TOADER 1. Introduction 2. Setting of the problem 3. Shape differentiation
More informationA finite element level set method for anisotropic mean curvature flow with space dependent weight
A finite element level set method for anisotropic mean curvature flow with space dependent weight Klaus Deckelnick and Gerhard Dziuk Centre for Mathematical Analysis and Its Applications, School of Mathematical
More informationAsymmetric Cheeger cut and application to multi-class unsupervised clustering
Asymmetric Cheeger cut and application to multi-class unsupervised clustering Xavier Bresson Thomas Laurent April 8, 0 Abstract Cheeger cut has recently been shown to provide excellent classification results
More informationKey words. Image segmentation, Euler s elastica, level set methods, Mumford-Shah segmentation model, additive model, overlapping objects.
A VARIATIONAL MODEL FOR SEGMENTATION OF OVERLAPPING OBJECTS WITH ADDITIVE INTENSITY VALUE YAN NEI LAW, HWEE KUAN LEE, CHAOQIANG LIU, AND ANDY M. YIP Abstract. We propose a variant of the Mumford-Shah model
More informationWritten Examination
Division of Scientific Computing Department of Information Technology Uppsala University Optimization Written Examination 202-2-20 Time: 4:00-9:00 Allowed Tools: Pocket Calculator, one A4 paper with notes
More informationExtending Continuous Cuts: Anisotropic Metrics and Expansion Moves
Extending Continuous Cuts: Anisotropic Metrics and Expansion Moves Carl Olsson, Martin Byröd, Niels Chr. Overgaard, Fredrik Kahl Centre for Mathematical Sciences, Lund University, Sweden P.O. Box 118,
More informationAccelerated Dual Gradient-Based Methods for Total Variation Image Denoising/Deblurring Problems (and other Inverse Problems)
Accelerated Dual Gradient-Based Methods for Total Variation Image Denoising/Deblurring Problems (and other Inverse Problems) Donghwan Kim and Jeffrey A. Fessler EECS Department, University of Michigan
More informationFoliations of hyperbolic space by constant mean curvature surfaces sharing ideal boundary
Foliations of hyperbolic space by constant mean curvature surfaces sharing ideal boundary David Chopp and John A. Velling December 1, 2003 Abstract Let γ be a Jordan curve in S 2, considered as the ideal
More informationOptimality Conditions for Constrained Optimization
72 CHAPTER 7 Optimality Conditions for Constrained Optimization 1. First Order Conditions In this section we consider first order optimality conditions for the constrained problem P : minimize f 0 (x)
More informationAnisotropic Total Variation Regularized L 1 -Approximation and Denoising/Deblurring of 2D Bar Codes
Anisotropic Total Variation Regularized L 1 -Approximation and Denoising/Deblurring of 2D Bar Codes Rustum Choksi Yves van Gennip Adam Oberman March 23, 2011 Abstract We consider variations of the Rudin-Osher-Fatemi
More informationA Linearly Convergent First-order Algorithm for Total Variation Minimization in Image Processing
A Linearly Convergent First-order Algorithm for Total Variation Minimization in Image Processing Cong D. Dang Kaiyu Dai Guanghui Lan October 9, 0 Abstract We introduce a new formulation for total variation
More informationUniversität des Saarlandes. Fachrichtung 6.1 Mathematik
Universität des Saarlandes U N I V E R S I T A S S A R A V I E N I S S Fachrichtung 6.1 Mathematik Preprint Nr. 402 Iterative TV-regularization of Grey-scale Images Martin Fuchs and Joachim Weickert Saarbrücken
More informationA level set method for solving free boundary problems associated with obstacles
A level set method for solving free boundary problems associated with obstacles K. Majava University of Jyväskylä, Department of Mathematical Information Technology, P.O. Box 35 (Agora), FIN-44 University
More informationOutline Introduction Edge Detection A t c i ti ve C Contours
Edge Detection and Active Contours Elsa Angelini Department TSI, Telecom ParisTech elsa.angelini@telecom-paristech.fr 2008 Outline Introduction Edge Detection Active Contours Introduction The segmentation
More informationA Four-Pixel Scheme for Singular Differential Equations
A Four-Pixel Scheme for Singular Differential Equations Martin Welk 1, Joachim Weickert 1, and Gabriele Steidl 1 Mathematical Image Analysis Group Faculty of Mathematics and Computer Science, Bldg. 7 Saarland
More informationNonlinear Flows for Displacement Correction and Applications in Tomography
Nonlinear Flows for Displacement Correction and Applications in Tomography Guozhi Dong 1 and Otmar Scherzer 1,2 1 Computational Science Center, University of Vienna, Oskar-Morgenstern-Platz 1, 1090 Wien,
More informationNonlinear Diffusion. Journal Club Presentation. Xiaowei Zhou
1 / 41 Journal Club Presentation Xiaowei Zhou Department of Electronic and Computer Engineering The Hong Kong University of Science and Technology 2009-12-11 2 / 41 Outline 1 Motivation Diffusion process
More informationLecture notes for Math 221, Winter 2013
Lecture notes for Math 221, Winter 2013 Lenya Ryzhik February 28, 2013 Essentially nothing found here is original except for a few mistakes and misprints here and there. These lecture notes are based on
More informationA MULTISCALE APPROACH IN TOPOLOGY OPTIMIZATION
1 A MULTISCALE APPROACH IN TOPOLOGY OPTIMIZATION Grégoire ALLAIRE CMAP, Ecole Polytechnique The most recent results were obtained in collaboration with F. de Gournay, F. Jouve, O. Pantz, A.-M. Toader.
More informationErkut Erdem. Hacettepe University February 24 th, Linear Diffusion 1. 2 Appendix - The Calculus of Variations 5.
LINEAR DIFFUSION Erkut Erdem Hacettepe University February 24 th, 2012 CONTENTS 1 Linear Diffusion 1 2 Appendix - The Calculus of Variations 5 References 6 1 LINEAR DIFFUSION The linear diffusion (heat)
More informationRisk Averse Shape Optimization
Risk Averse Shape Optimization Sergio Conti 2 Martin Pach 1 Martin Rumpf 2 Rüdiger Schultz 1 1 Department of Mathematics University Duisburg-Essen 2 Rheinische Friedrich-Wilhelms-Universität Bonn Workshop
More informationLINEARIZED BREGMAN ITERATIONS FOR FRAME-BASED IMAGE DEBLURRING
LINEARIZED BREGMAN ITERATIONS FOR FRAME-BASED IMAGE DEBLURRING JIAN-FENG CAI, STANLEY OSHER, AND ZUOWEI SHEN Abstract. Real images usually have sparse approximations under some tight frame systems derived
More informationThe Skorokhod reflection problem for functions with discontinuities (contractive case)
The Skorokhod reflection problem for functions with discontinuities (contractive case) TAKIS KONSTANTOPOULOS Univ. of Texas at Austin Revised March 1999 Abstract Basic properties of the Skorokhod reflection
More informationKey words: denoising, higher-order regularization, stability, weak convergence, Brezis-Lieb condition. AMS subject classifications: 49N45, 94A08.
A FEW REMARKS ON VARIATIONAL MODELS FOR DENOISING RUSTUM CHOKSI, IRENE FONSECA, AND BARBARA ZWICKNAGL Abstract. Variational models for image and signal denoising are based on the minimization of energy
More informationError Analysis for H 1 Based Wavelet Interpolations
Error Analysis for H 1 Based Wavelet Interpolations Tony F. Chan Hao-Min Zhou Tie Zhou Abstract We rigorously study the error bound for the H 1 wavelet interpolation problem, which aims to recover missing
More informationPROBABILISTIC CONTINUOUS EDGE DETECTION USING LOCAL SYMMETRY
PROBABILISTIC CONTINUOUS EDGE DETECTION USING LOCAL SYMMETRY Gerald Mwangi, Paul Fieguth, Christoph S. Garbe : University of Heidelberg, Germany, : University of Waterloo, Canada ABSTRACT We describe a
More informationPDE-based image restoration, I: Anti-staircasing and anti-diffusion
PDE-based image restoration, I: Anti-staircasing and anti-diffusion Kisee Joo and Seongjai Kim May 16, 2003 Abstract This article is concerned with simulation issues arising in the PDE-based image restoration
More informationREGULAR LAGRANGE MULTIPLIERS FOR CONTROL PROBLEMS WITH MIXED POINTWISE CONTROL-STATE CONSTRAINTS
REGULAR LAGRANGE MULTIPLIERS FOR CONTROL PROBLEMS WITH MIXED POINTWISE CONTROL-STATE CONSTRAINTS fredi tröltzsch 1 Abstract. A class of quadratic optimization problems in Hilbert spaces is considered,
More informationNonlinear regularizations of TV based PDEs for image processing
Nonlinear regularizations of TV based PDEs for image processing Andrea Bertozzi, John Greer, Stanley Osher, and Kevin Vixie ABSTRACT. We consider both second order and fourth order TV-based PDEs for image
More informationA Necessary and Sufficient Condition for the Continuity of Local Minima of Parabolic Variational Integrals with Linear Growth
A Necessary and Sufficient Condition for the Continuity of Local Minima of Parabolic Variational Integrals with Linear Growth E. DiBenedetto 1 U. Gianazza 2 C. Klaus 1 1 Vanderbilt University, USA 2 Università
More informationLEVEL SET BASED MULTISPECTRAL SEGMENTATION WITH CORNERS
LEVEL SET BASED MULTISPECTRAL SEGMENTATION WITH CORNERS WENHUA GAO AND ANDREA BERTOZZI Abstract. In this paper we propose an active contour model for segmentation based on the Chan-Vese model. The new
More informationOn a weighted total variation minimization problem
On a weighted total variation minimization problem Guillaume Carlier CEREMADE Université Paris Dauphine carlier@ceremade.dauphine.fr Myriam Comte Laboratoire Jacques-Louis Lions, Université Pierre et Marie
More informationGlobal minimization. Chapter Upper and lower bounds
Chapter 1 Global minimization The issues related to the behavior of global minimization problems along a sequence of functionals F are by now well understood, and mainly rely on the concept of -limit.
More informationMulti-Robotic Systems
CHAPTER 9 Multi-Robotic Systems The topic of multi-robotic systems is quite popular now. It is believed that such systems can have the following benefits: Improved performance ( winning by numbers ) Distributed
More informationNear-Potential Games: Geometry and Dynamics
Near-Potential Games: Geometry and Dynamics Ozan Candogan, Asuman Ozdaglar and Pablo A. Parrilo January 29, 2012 Abstract Potential games are a special class of games for which many adaptive user dynamics
More informationOn Nonlinear Dirichlet Neumann Algorithms for Jumping Nonlinearities
On Nonlinear Dirichlet Neumann Algorithms for Jumping Nonlinearities Heiko Berninger, Ralf Kornhuber, and Oliver Sander FU Berlin, FB Mathematik und Informatik (http://www.math.fu-berlin.de/rd/we-02/numerik/)
More informationInterior-Point Methods for Linear Optimization
Interior-Point Methods for Linear Optimization Robert M. Freund and Jorge Vera March, 204 c 204 Robert M. Freund and Jorge Vera. All rights reserved. Linear Optimization with a Logarithmic Barrier Function
More informationVariational methods for restoration of phase or orientation data
Variational methods for restoration of phase or orientation data Martin Storath joint works with Laurent Demaret, Michael Unser, Andreas Weinmann Image Analysis and Learning Group Universität Heidelberg
More informationLecture 4 Colorization and Segmentation
Lecture 4 Colorization and Segmentation Summer School Mathematics in Imaging Science University of Bologna, Itay June 1st 2018 Friday 11:15-13:15 Sung Ha Kang School of Mathematics Georgia Institute of
More informationCONSTRAINED PERCOLATION ON Z 2
CONSTRAINED PERCOLATION ON Z 2 ZHONGYANG LI Abstract. We study a constrained percolation process on Z 2, and prove the almost sure nonexistence of infinite clusters and contours for a large class of probability
More informationA Posteriori Estimates for Cost Functionals of Optimal Control Problems
A Posteriori Estimates for Cost Functionals of Optimal Control Problems Alexandra Gaevskaya, Ronald H.W. Hoppe,2 and Sergey Repin 3 Institute of Mathematics, Universität Augsburg, D-8659 Augsburg, Germany
More informationSimulated Annealing for Constrained Global Optimization
Monte Carlo Methods for Computation and Optimization Final Presentation Simulated Annealing for Constrained Global Optimization H. Edwin Romeijn & Robert L.Smith (1994) Presented by Ariel Schwartz Objective
More information1/12/05: sec 3.1 and my article: How good is the Lebesgue measure?, Math. Intelligencer 11(2) (1989),
Real Analysis 2, Math 651, Spring 2005 April 26, 2005 1 Real Analysis 2, Math 651, Spring 2005 Krzysztof Chris Ciesielski 1/12/05: sec 3.1 and my article: How good is the Lebesgue measure?, Math. Intelligencer
More informationThe Deep Ritz method: A deep learning-based numerical algorithm for solving variational problems
The Deep Ritz method: A deep learning-based numerical algorithm for solving variational problems Weinan E 1 and Bing Yu 2 arxiv:1710.00211v1 [cs.lg] 30 Sep 2017 1 The Beijing Institute of Big Data Research,
More informationA VARIATIONAL MODEL FOR INFINITE PERIMETER SEGMENTATIONS BASED ON LIPSCHITZ LEVEL SET FUNCTIONS: DENOISING WHILE KEEPING FINELY OSCILLATORY BOUNDARIES
A VARIATIONAL MODEL FOR INFINITE PERIMETER SEGMENTATIONS BASED ON LIPSCHITZ LEVEL SET FUNCTIONS: DENOISING WHILE KEEPING FINELY OSCILLATORY BOUNDARIES MARCO BARCHIESI, SUNG HA KANG, TRIET M. LE, MASSIMILIANO
More informationIntroductory Analysis I Fall 2014 Homework #9 Due: Wednesday, November 19
Introductory Analysis I Fall 204 Homework #9 Due: Wednesday, November 9 Here is an easy one, to serve as warmup Assume M is a compact metric space and N is a metric space Assume that f n : M N for each
More informationUnconstrained optimization
Chapter 4 Unconstrained optimization An unconstrained optimization problem takes the form min x Rnf(x) (4.1) for a target functional (also called objective function) f : R n R. In this chapter and throughout
More informationTWO-PHASE APPROACH FOR DEBLURRING IMAGES CORRUPTED BY IMPULSE PLUS GAUSSIAN NOISE. Jian-Feng Cai. Raymond H. Chan. Mila Nikolova
Manuscript submitted to Website: http://aimsciences.org AIMS Journals Volume 00, Number 0, Xxxx XXXX pp. 000 000 TWO-PHASE APPROACH FOR DEBLURRING IMAGES CORRUPTED BY IMPULSE PLUS GAUSSIAN NOISE Jian-Feng
More informationMathematical Imaging with Applications to MRI and Diffusion Tensor MRI. Johan Lie. PhD Thesis. Department of Mathematics University of Bergen
Mathematical Imaging with Applications to MRI and Diffusion Tensor MRI PhD Thesis Johan Lie Department of Mathematics University of Bergen January 2007 Preface This dissertation is submitted as a partial
More informationSolution-driven Adaptive Total Variation Regularization
1/15 Solution-driven Adaptive Total Variation Regularization Frank Lenzen 1, Jan Lellmann 2, Florian Becker 1, Stefania Petra 1, Johannes Berger 1, Christoph Schnörr 1 1 Heidelberg Collaboratory for Image
More informationANALYSIS OF THE TV REGULARIZATION AND H 1 FIDELITY MODEL FOR DECOMPOSING AN IMAGE INTO CARTOON PLUS TEXTURE. C.M. Elliott and S.A.
COMMUNICATIONS ON Website: http://aimsciences.org PURE AND APPLIED ANALYSIS Volume 6, Number 4, December 27 pp. 917 936 ANALYSIS OF THE TV REGULARIZATION AND H 1 FIDELITY MODEL FOR DECOMPOSING AN IMAGE
More informationNear-Potential Games: Geometry and Dynamics
Near-Potential Games: Geometry and Dynamics Ozan Candogan, Asuman Ozdaglar and Pablo A. Parrilo September 6, 2011 Abstract Potential games are a special class of games for which many adaptive user dynamics
More informationHamburger Beiträge zur Angewandten Mathematik
Hamburger Beiträge zur Angewandten Mathematik Numerical analysis of a control and state constrained elliptic control problem with piecewise constant control approximations Klaus Deckelnick and Michael
More informationGeneralized Newton-Type Method for Energy Formulations in Image Processing
Generalized Newton-Type Method for Energy Formulations in Image Processing Leah Bar and Guillermo Sapiro Department of Electrical and Computer Engineering University of Minnesota Outline Optimization in
More informationarxiv: v1 [math.st] 1 Dec 2014
HOW TO MONITOR AND MITIGATE STAIR-CASING IN L TREND FILTERING Cristian R. Rojas and Bo Wahlberg Department of Automatic Control and ACCESS Linnaeus Centre School of Electrical Engineering, KTH Royal Institute
More informationMathematical Problems in Image Processing
Gilles Aubert Pierre Kornprobst Mathematical Problems in Image Processing Partial Differential Equations and the Calculus of Variations Second Edition Springer Foreword Preface to the Second Edition Preface
More information3.10 Lagrangian relaxation
3.10 Lagrangian relaxation Consider a generic ILP problem min {c t x : Ax b, Dx d, x Z n } with integer coefficients. Suppose Dx d are the complicating constraints. Often the linear relaxation and the
More informationAppendix A Taylor Approximations and Definite Matrices
Appendix A Taylor Approximations and Definite Matrices Taylor approximations provide an easy way to approximate a function as a polynomial, using the derivatives of the function. We know, from elementary
More informationAn improved active contour model based on level set method
1 015 1 ( ) Journal of East China Normal University (Natural Science) No. 1 Jan. 015 : 1000-5641(015)01-0161-11, (, 0006) :,., (SPF),.,,.,,,,. : ; ; ; : O948 : A DOI: 10.3969/j.issn.1000-5641.015.01.00
More informationRestoration of images with rotated shapes
Numer Algor (008) 48:49 66 DOI 0.007/s075-008-98-y ORIGINAL PAPER Restoration of images with rotated shapes S. Setzer G. Steidl T. Teuber Received: 6 September 007 / Accepted: 3 January 008 / Published
More informationInvestigating the Influence of Box-Constraints on the Solution of a Total Variation Model via an Efficient Primal-Dual Method
Article Investigating the Influence of Box-Constraints on the Solution of a Total Variation Model via an Efficient Primal-Dual Method Andreas Langer Department of Mathematics, University of Stuttgart,
More information1.The anisotropic plate model
Pré-Publicações do Departamento de Matemática Universidade de Coimbra Preprint Number 6 OPTIMAL CONTROL OF PIEZOELECTRIC ANISOTROPIC PLATES ISABEL NARRA FIGUEIREDO AND GEORG STADLER ABSTRACT: This paper
More informationc 2011 International Press Vol. 18, No. 1, pp , March DENNIS TREDE
METHODS AND APPLICATIONS OF ANALYSIS. c 2011 International Press Vol. 18, No. 1, pp. 105 110, March 2011 007 EXACT SUPPORT RECOVERY FOR LINEAR INVERSE PROBLEMS WITH SPARSITY CONSTRAINTS DENNIS TREDE Abstract.
More informationComputers and Mathematics with Applications
Computers and Mathematics with Applications 6 () 77 745 Contents lists available at ScienceDirect Computers and Mathematics with Applications journal homepage: www.elsevier.com/locate/camwa Multiphase
More informationIMAGE RESTORATION: TOTAL VARIATION, WAVELET FRAMES, AND BEYOND
IMAGE RESTORATION: TOTAL VARIATION, WAVELET FRAMES, AND BEYOND JIAN-FENG CAI, BIN DONG, STANLEY OSHER, AND ZUOWEI SHEN Abstract. The variational techniques (e.g., the total variation based method []) are
More informationA primal-dual approach for a total variation Wasserstein flow
A primal-dual approach for a total variation Wasserstein flow Martin Benning 1, Luca Calatroni 2, Bertram Düring 3, Carola-Bibiane Schönlieb 4 1 Magnetic Resonance Research Centre, University of Cambridge,
More informationRobust error estimates for regularization and discretization of bang-bang control problems
Robust error estimates for regularization and discretization of bang-bang control problems Daniel Wachsmuth September 2, 205 Abstract We investigate the simultaneous regularization and discretization of
More informationThe BV space in variational and evolution problems
The BV space in variational and evolution problems Piotr Rybka the University of Warsaw and the University of Tokyo rybka@mimuw.edu.pl November 29, 2017 1 The definition and basic properties of space BV
More informationOn Friedrichs inequality, Helmholtz decomposition, vector potentials, and the div-curl lemma. Ben Schweizer 1
On Friedrichs inequality, Helmholtz decomposition, vector potentials, and the div-curl lemma Ben Schweizer 1 January 16, 2017 Abstract: We study connections between four different types of results that
More informationStationary mean-field games Diogo A. Gomes
Stationary mean-field games Diogo A. Gomes We consider is the periodic stationary MFG, { ɛ u + Du 2 2 + V (x) = g(m) + H ɛ m div(mdu) = 0, (1) where the unknowns are u : T d R, m : T d R, with m 0 and
More informationNewton Method with Adaptive Step-Size for Under-Determined Systems of Equations
Newton Method with Adaptive Step-Size for Under-Determined Systems of Equations Boris T. Polyak Andrey A. Tremba V.A. Trapeznikov Institute of Control Sciences RAS, Moscow, Russia Profsoyuznaya, 65, 117997
More information5 Handling Constraints
5 Handling Constraints Engineering design optimization problems are very rarely unconstrained. Moreover, the constraints that appear in these problems are typically nonlinear. This motivates our interest
More informationOn an Approximation Result for Piecewise Polynomial Functions. O. Karakashian
BULLETIN OF THE GREE MATHEMATICAL SOCIETY Volume 57, 010 (1 7) On an Approximation Result for Piecewise Polynomial Functions O. arakashian Abstract We provide a new approach for proving approximation results
More informationCoordinate descent. Geoff Gordon & Ryan Tibshirani Optimization /
Coordinate descent Geoff Gordon & Ryan Tibshirani Optimization 10-725 / 36-725 1 Adding to the toolbox, with stats and ML in mind We ve seen several general and useful minimization tools First-order methods
More informationCalculus of Variations Summer Term 2014
Calculus of Variations Summer Term 2014 Lecture 20 18. Juli 2014 c Daria Apushkinskaya 2014 () Calculus of variations lecture 20 18. Juli 2014 1 / 20 Purpose of Lesson Purpose of Lesson: To discuss several
More information